Nom original: mémoire.pdf
Ce document au format PDF 1.5 a été généré par LaTeX with hyperref package / pdfTeX-1.40.18, et a été envoyé sur fichier-pdf.fr le 10/02/2018 à 18:04, depuis l'adresse IP 88.174.x.x.
La présente page de téléchargement du fichier a été vue 168 fois.
Taille du document: 351 Ko (21 pages).
Confidentialité: fichier public
Télécharger le fichier (PDF)
Aperçu du document
INFLUENCE OF SOCIOECONOMIC
FACTORS ON THE OVERALL STATE
OF HEALTH IN CALIFORNIA
Master’s Thesis in Applied Econometrics
Ouriane A¨ıssou, Yasmine Bunout, Benjamin Iz´erable
Supervised by H´el`ene Huber∗and Philippe De Peretti†
University of Paris 1 Panth´eon-Sorbonne
Master’s degree in Econometrics and Statistics (MoSEF)
This article estimates the effect of socioeconomic determinants over
health using californian cross section data for the year 2016. We found
that variables representing large population like skin color or gender, tend
to have less effect on health than more personal determinants like education or the working status. Eventually, what this study shows is that
health is not perfectly link to one’s income but also to his social environment and that this environment play a major hidden role.
keywords : Health, Education, Socioeconomic impact, California.
JEL Classification : C21, C51, I18, I20.
∗ Associate Professor in Economics, University of Paris 1 Panth´
eon-Sorbonne and Paris
School of Economics. Member of Health, Risk, Insurance Chair of University of ParisDauphine.
† Associate Professor in Economics, University of Paris 1 Panth´
eon-Sorbonne. Member of
the Council of The Society for Economic Measurement and Associate Editor of the Review of
Finance and Banking.
2 Literature review
2.1 The production of health . . . . . . . . . . . . . . . . . . .
2.2 The role of socioeconomic factors . . . . . . . . . . . . . . .
3.1 Presentation of the dataset . .
3.1.1 Design . . . . . . . . . .
3.1.2 Outcome measure . . .
3.2 Model . . . . . . . . . . . . . .
3.2.1 Econometric process . .
3.2.2 Formal approach . . . .
3.2.3 The odds ratio method
4.1 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2 Interpretations . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.1 The dataset . . . . . . . . . . . . . . . . . . . .
5.1.1 The theoretical cross-section constraint
5.1.2 The problem of the outcome measure .
5.2 The model . . . . . . . . . . . . . . . . . . . . .
5.3 Suggestions . . . . . . . . . . . . . . . . . . . .
”There is a remarkably close link between where you are on the socioeconomic ladder and your health – the higher the rank, the better the health.” 1
It is a well-known fact that income inequalities are very consequent in the
United States of America. Nowadays, there is a growing number of people in the United States suffering from being unable to get health care,
which leads them to make a strict arbitration regarding health. Indeed,
social status and income play a vital role in access to health care. Indeed,
income inequalities cause all kind of other inequalities such as medical access. This inequality of medical access leads inescapably to a degradation
in state of health. As a result, we decide to analyze the socioeconomic impacts on the overall state of health in the United States of America (USA).
There are two separate insurances in the United States : private insurance
that is offered by companies financed by social contributions and public
insurance that covers disadvantaged population and old age. However,
over 45 million people in 2009 are uninsured, which is explained by their
relatively low salary level which does not allow them, on the one hand to
be insured by their employers and on the other hand not to benefit from
public insurance. This widens the inequality in the United States, moreover this climbing figure is due to a rise in insurance premiums which
is not followed by a rise in wages (Galvis-Narinos F., Mont´elmard A.,
2009) . Nevertheless, despite the high level of technology and quality
of care, distribution remains highly unequal and practices very expensive.
Galvinos-Narinos and Mont´elimard (2009) explain that the high price of
general practitioners consultations can be explained by the fact that they
have more expenses than the French doctors. Indeed, higher education
in the United States is very expensive, which leads students to take out
loans that they will have to repay throughout their lives. This leads
them to a desire to value their remuneration but on the other hand this
questions us about the real quality of care. Nevertheless, income is certainly an essential factor in the cause of health differences, but we believe
that there are other factors influencing health status. Indeed, the transmission of a patrimony on the part of parents, such as culture, could lead
to differences of knowledge concerning the various possibilities open to us.
The aim of this thesis is to demonstrate that socioeconomic factors can
clearly influence the overall health state of individuals. In order to carry
out our analysis we will base ourselves on a californian survey covering
a great diversity of people. Our purpose is to determine which socioeconomic factors influence the overall health status. We want to know if the
state of health changes according to social affiliation, age or ethnicity. In
order to model this, we use the logistic regression so that the dependant
variable represents the probability to be in poor health in interactions
with socioeconomic factors. As it is explained in the program Healthy
People 2020, one know that health is firstly influenced by the close en1 ‘The richer you are, the better your health – and how this can be changed’, Michael
Marmot. The Guardian, September 2015.
vironment, alimentary habits and surroundings. Thus, we will focus our
study here on qualitative variables that represent these dimensions.
The production of health
The point of view of economists could appear quite particular at first sight
concerning health’s analysis. Actually, the economic approach assimilates
health as an asset that can be produced, so as the producer who maximize its profit. But here, the individual maximizes its health production.
This process has been chosen in the Grossman Model (1972) . In the
model, the rational individual maximizes its utility over two periods by
maximizing its stock of health and consumption. The particularity of
these states of health is that both are valued as a stock of health -in the
same way as a stock of capital. The utility is affected by a variable that
represents the amount of sick time which is the lower the larger the stock
of health is. The second period depends on an amount of ”investment”
that can be increased by the purchase of medical services or spendings in
units of time on preventive effort. Wagstaff in 1993  has proposed to
go into the Grossman Model in depth in introducing the variable education to explain the corresponding investment function. Hence, Wagstaff
deduces the structural demand for medical services, where a high level of
education lower the demand for medical services in such way that a high
level of education increases the productivity of medical services -by the
fact that highly educated people have more information on health risks
and are more prompt to follow indications proposed by physicians. Also,
in derivating a given utility function, Wagstaff shows somehow that the
higher the level of education is, the lower the marginal cost of health will
be. Moreover, the initial wage rate increases the demand for medical services for a given level of education because it facilitates access to health
information. Eventually, this theoretical model gives some intuitions for
empirical approaches, notably on the role of some socioeconomic determinants, such as the level of education or wage, on health status.
Socioeconomic factors influence the state of health of individuals. Nonetheless, the effectiveness of medicine also has a major role in the state of
health in a society. Auster et al.  in 1969 proposed to analyze the
effectiveness of medicine by explaining a standardized mortality rate with
socioeconomic, medical and purely economics factors. With this aim in
mind, they estimate correspondents parameters with the Ordinary least
squares (OLS) method with a cross-section dataset of 48 States in the
USA during 1960. In order to do this, the mortality rate adopt the economic concept of Cobb-Douglas production function. Thus, they found
negative elasticities of -0.051 for the pharmaceutical outlay per capita and
of -0.190 for the medical auxiliary staff per capita, ceteris paribus. The
existence of a medical school also seems to contribute to the reduction of
the mortality rate (-0.034), ceteris paribus. Actually, it makes sense because the higher the offer of health services, the lower the mortality rate.
On the other hand, the share of industry in total employment and the
number of physicians per capita seem to contribute positively on the mor-
tality rate (respectively for 0.051 and 0.143), ceteris paribus 2 . If in a first
time, results for the share of industry in the total employment. Actually,
the second results appears counterintuitive because we can believe that
the higher the number of physicians per capita, the lower the mortality
rate. Nonetheless it appears that it is not the case. Actually, despite this
counter-intuitive logic, this could be explained by the fact that, in this
case, the demand could be induced by the offer. This phenomenon it is
named Supplier-Induced Demand theory. It results from the asymmetric
information whose suffers the patient and it is peculiar to Health Economics. The physician uses its superior information in order to encourage
the patient to demand a greater quantity of healthcare services .
The role of socioeconomic factors
Subramanian, Kim and Kawachi’s article (2005)  on the co-variation in
the socioeconomic determinants of self rated health and happiness, shows
that happiness leads to better health. Indeed, in order to demonstrate
their results, they have decided to conduct a multivariate analysis of individuals and communities in the United States of America. The most
conclusive results and the ones that interested us the most in our analysis
are the variables Race, Education and Income. Indeed, they have demonstrated that when we were not white, we were 25 percent more likely of
being in poor health and therefore, unhappy. What is interesting is that
these variables are the strongest of the model; they could be a strong
determinant of the state of health. In addition, they have also concluded
that when an individual had a low level of education, he was 3.57 times
more likely to be in poor health; moreover, as health is correlated with
happiness, the authors found that these people were 2.68 times unhappier
than people with a better education. One of the other variables we will encounter in our model is Income. Very poor people are 4 times more likely
to be in poor health, which is in agreement with the American healthcare
system. These individuals will not be able to finance their medical expenses.
Other articles relate more particularly to the inequalities of access to
health as well as the renunciation of healthcare for financial reasons. Indeed, since we don’t all have the same income, we are not able to benefit
from the same quality of healthcare. This creates an inequality that needs
to be further considered in order to provide a better distribution of health
care. The econometric approach on the renunciation of access to care
for financial reasons led by the IRDES3 (Despres, Dourgnon, 2011) ,
has shown that the most explanatory factor of the renunciation is social belonging. In order to carry out their analysis, their database was
constructed from the Health and Protection Survey (ESP) based on questions about health, social and economic situation of individuals as well
as on their choice of social insurance. It should also be noted that the
2 In this article, other results have been found as the impact of education or of the income,
nonetheless they are not significant.
3 French Institute of research and documentation in health economics (IRDES).
results showed that renunciation was more present for dental care. In addition, this article allowed us to confirm that socioeconomic factors have
a long-term impact on our health care consumption decisions. Despite
the fact that the french healthcare system is more egalitarian than the
American system, there are nevertheless many inequalities between social
classes. An individual who has only a subsistence income will be more
constrained than an individual receiving a higher salary, leading the latter
to make a real arbitration.
Furthermore, financial resources are certainly deeply correlated with state
of health but a high level of education tends towards better financial resources. In fact, for Nicholas Freudengerg and Jessica Ruglis (2007), ”robust epidemiological evidence suggests that education is such an elixir” 
because it clearly leads to a better health. This report claims that ”education is one of the strongest predictor of health” as well. Conversely, they
turn over the problem and insist on the necessity to reduce school dropout.
Then health interventions go hand in hand with education interventions.
Both are endogenous because a good health lead to a good education
and vice versa. In this article, authors suggest several recommendations
for public authority. The first is to reduce the education failure in high
school and the second is to improve research on reciprocal connections
between education and health. In fact, they propose to combine both
interventions, notably at school with improving education conditions and
investing in prevention program.
Presentation of the dataset
The analysis was based on the 2016 California Health Interview Survey
(CHIS 2016), developed by the centre of health policy research from the
University of California in collaboration with the California Department
of Public Health and the Department of Health Care Services, it is the
nation’s largest state-level health survey. The aim of this survey is to
analyze public health’s access and health care needs in order to advocate
policies to find better financing alternatives to coverage and access to
health care. This dataset includes a sample of 20 000 individual observations which were collected between January and December of 2016. The
CHIS has collected an extensive information for all age groups on health
status, health condition, health related behaviours, health insurance coverage, access to health care services, and other health and health-related
issues. Respondent had to choose an answer among predefined choices.
Moreover, the CHIS 2016 chose to use a data collection according to a
population based on a telephone survey of California using random digit
dialing (RDD). In order to update the precedents surveys, and to take
into account the major part of the population, they have decided to add
a gender identity and to take into account 6 languages (English, Spanish,
Chinese, Vietnamese, Korean and Tagalog). Indeed, the CHIS 2016, is
the first survey that asks questions to identify transgender or other gender identity. In addition, they relied on the following characteristics to
obtain the most complete database possible: medical eligibility measure,
health insurance measure, cell phone sample, race and ethnicity coding.
Plus, concerning the missing data/responses, they have weighted the sample in order to compensate the biases created because of the non-respondents
(who may have different characteristics than the others). Furthermore,
concerning the imputation method they have used two different methods:
a hot deck imputation and a data editing that put a valid replacement
value based on a known value of the other variables of the same responses.
Then the sample is calibrated for representing the most accurately the all
population of California according to the organism. So there is no problem
Design of the survey. The respondents had to fulfill a survey. For
instance, concerning the case of the state of health -that represents the
dependant variable of the model-, the respondent had to choose an answer
among several possibilities. Then, in this case, they had to reply to the
question ”Would you say that in general your health is excellent, very
good, good, fair or poor ? ”. Hence, the respondent have to choose among
the following propositions: Excellent; very good ; good ; fair ; poor ; refused ;
don’t know. As a result, the totality of variables are calibrated as a dummy.
About the dependant variable. In order to consider the state of
health as the dependant variable, we had to transform this dummy as a
binary variable. So we have made the choice to consider that an individual
is in poor health when one replied fair and poor. Thus, individuals who
have replied good, very good and excellent are considered as in good health.
A logit model. The purpose of the model consists on observing if
general socioeconomic factors can explain inequalities in health state of
individuals. In other terms, we want to know if the health state of individuals can be linked with the social class in the Californian society. Furthermore, the totality of independent variables that we use are dummy.
Then it appears that a good way to model a probability for being in poor
health in interaction with independent dummy variables is to use the logit
model based on polytomics. So all independent variables have their variable reference due to the polytomic nature. Thus, the dependant variable
Yi of n individuals sub-scripted by i = 1, .., n can take two forms :
1 = if the individual is in poor health
0 = if the individual is in good health
E [Yi ] = P (Yi = 1) .
Our logit model is written in this way:
P (Yi = 1|X) = 1 + exp −βo + β1 usnati + β2 nonusi + β3 unempi
+ β4 tenanti + β5 notobesei + β6 nobachi + β7 f emi + β8 whitei +
β9 uvisiti + β10 cityi
Variable specifications. We firstly select a range of dummy variables that illustrate the citizenship of individuals. There is a distinction
between individuals who are native American citizen (usborni ), from those
who are naturalized American citizen (natusi ) and those who are nonAmerican citizen (nonusi ). The variable usborni is classed as reference.
We secondly have chosen to see if the fact to live in city (cityi ) can affect
the personal state of health compared to not live in city (nocityi classed as
reference). In this model, we also want to analyze if the work status can
affect the health of individuals. There are two variables illustrating this.
Then, either the person is employed (empi ) nor is unemployed (unempi
classed as reference). The employed group includes persons who have
full-time and partial-time work status. The unemployed group includes
individuals who are looking for a job4 and who are not looking for a job.
In addition, we want to know if the fact of being an owner can have an
impact on the state of health. Then we introduce a dummy variable that
shows if the individual is an owner (owni ) or a tenant (tenanti took as
it represents a very little part of the sample.
reference). Moreover it can be interesting to analyze if the degree of the
individual can have an impact on his state of health. Thus, we selected
a range of dummy variables that mention the degree of the individual.
That variable has been coded in such way that either the individual has
a bachelor’s degree (bachi took as reference) or more nor has less than
a bachelor’s degree (nobachi ). Furthermore we introduce in our model
some health dummy variable ranges. The first range is about the obesity
because it has clearly an effect in the state of health. Then it differentiates the obese person (obesei ) with the non-obese (notobesei classed as
reference). The second range is about medical visits. It also differentiates
a person who goes more than 4 times per year in medical visit (ovisiti
took as reference) and who goes less 3 times per year (uvisiti ). Eventually, last but not least, we want to observe if the ethnicity can influence
the personal state of health. Then let us introduce a dummy range of
variables that differentiates the white person (whitei took as reference)
with the non-white person (whitei ).
Here the logit model allows us to analyze the occurrence of the considered
event in accordance with observed characteristics (Xi1 , .., Xik ) that we
will specify later. In this model, there is a combination5 of explanatory
variables6 Xi = (1, Xi1 , .., Xik ) with parameters β = (β0 , β1 , ...βk )0 where
Yi = Xi β + with the residual term . In this case, we want to explain
the value of Y with X, or more precisely, to estimate P (Yi = 1|Xi ). So
the dichotomic variable follows a Bernoulli distribution of p parameter.
p (1 − p)1−0 = 1 − p if k = 0
P (Yi = k) = p (1 − p)
p1 (1 − p)1−1 = p
if k = 1
For using the logistic regression, we need to assume these assumptions :
Assumption 1. is classed as a random variable of symmetric distribution.
Assumption 2. There is no endogeneity. In other words is independent
from Xi .
Nonetheless, notice that the second assumption is very strong because
there is endogeneity with these explanatory variables. For instance the
revenue is endogeneous with the weight or medical visits. But we need to
deal with it otherwise the model cannot be built. Then we have :
P (Yi = 1|Xi ) = P (Xi β + i ≥ 0|Xi ) P (Yi = Xi β ≥ − i |Xi ) = F− (Xi β)
Where F− (Xi β) = F (Xi β) because of the symmetry of the residual law.
The specificity of the logit model is that assumes logistic distribution
function as follows:
F (Xi β) =
1 + exp (−Xi β)
in parameters but not in variables.
that will be specify later.
In this case, our linear econometric model is formally written as follows:
yi = βo − β1 usnati − β2 nonusi − β3 unempi − β4 tenanti − β5 notobesei
− β6 nobachi − β7 f emi − β8 whitei − β9 uvisiti − β10 cityi + i
And its logistic formalization is given by:
P (Yi = 1|X) = 1 + exp −βo + β1 usnati + β2 nonusi + β3 unempi
+ β4 tenanti + β5 notobesei + β6 nobachi + β7 f emi + β8 whitei + β9 uvisiti
+ β10 cityi
With its density function:
f (Xi β) =
∂F (Xi β)
= F (Xi β) [1 − F (Xi β)] .
∂ (Xi β)
The estimation of parameters is given by the maximum likelihood method.
The likelihood is written:
L (Xi β) =
F (Xi β)Yi [1 − F (Xi β)]1−Yi
We deduce the log-likelihood:
ln L (Xi β) =
Yi ln F (Xi β) +
(1 − Yi ) ln [1 − F (Xi β)]
And its parameters (β) are given by finding these optimums
∂ ln L (Xi β)
∂ 2 ln L (Xi β)
The odds ratio method
Estimators found by the log-likelihood method cannot be used directly
because they are defined within a constant. Then, we use odds ratios φ.
For exemple, the odds ratio between X = 1 and X = 2 is written in this
P (Y = 1|X = 1) P (Y = 1|X = 2)
P (Y = 0|X = 1) P (Y = 0|X = 2)
In fact, this cannot be interpreted as an absolute probability but more as
a relative probability. For instance, if this ratio is equal to 2, then the
probability that X = 1 occur is two times more to occur than for X = 2.
practice, the calculation is made by the logistic procedure of SAS.
(ref = usborn)
(ref = nocity)
(ref = emp)
(ref = uvist)
(ref = own)
(ref = notobese)
(ref = bach)
(ref = white)
(ref = f em)
*** p-value < 0.001
** p-value < 0.005
* p-value < 0.010
N = 19906
βˆ −→ β
Table 1: Parameter estimations by the maximum of likelihood method for the
probability to be in poor health (α = 0.005)
Table 2: Test of global nullity : β = 0
Figure 1: The ROC Curve
Table 3: Odds ratio and Wald confidence interval. α = 0.005
Figure 2: Odds ratio intervals
According to results (table 3), the totality of our estimated parameters
are strongly significant except for gender and home localization variables.
Moreover, the test of global nullity (table 3) approves the significance of
the model for a threshold critical of 1%. That is to say literally that there
is at least one factor that influence the dependant variable. The high
number of observations (19 906) insures the convergence of the estimator,
so there is asymptotic results. Also, according to the area under the
ROC8 curve, (0.7491)9 this is a good model as it can quietly well separate
individuals who are in good health from individuals who are in poor health
according to the form of the curve.
Citizenship status. Odds ratio shows that a naturalized Americans is
1.823 times more likely to be in poor health relatively to an individual who
is born American citizen. And worse for the non-citizen. The citizenship
clearly impacts the state of health of individuals. In fact, it is quite logical
because generally the fact of being non-citizen is a sign of precarious
situation. Non-citizens have poor incomes and they often do not speak
the language of the country. These elements do not foster individuals
to access at medical care. Even worse in a country where the medical
care access framework is determined by market’s law. Logically, the effect
is a little reduced for the American-naturalized person but remains high
relatively to the someone who is born American citizen anyway.
= Receiver Operating Characteristic curve.
Figure 1 above.
Home localization. The results for Home localization variables do
not indicate a large difference between the fact of living in the city or
not. We did not expect that. Actually, we introduced this variable in the
belief that people who live in town could be more exposed to pollution
than people who do not. However, our model finally shows that people
who do not live in the city are 1.115 times more likely to be in poor
health than people who do. In reality, the odds ratio is low, so that the
difference between these two groups is not large. Nevertheless, there is still
a difference, and it can be due to two things : the first is that tendentiously,
people who do not live in town suffer from a lack of medical care supply;
the second is that people who live in the city have more revenues that
people who do not, so they have a higher access to medical care. But let’s
be careful about this according to the weak value of the corresponding
odds ratio and the significance of the parameter.
Working status. The working status plays a major part in the state
of health of individuals. Indeed, odds ratio shows that unemployed people
are 2.677 times more likely to be in poor health than people who work.
Actually, the difference is high, and it makes sense. People who have a job
are more likely to have a good health because having a job permit them
to be insured by their employers contrary to unemployed people so the
amount to be paid by the worker is much lower than what the unemployed
person has to pay. Even worse, the effect is magnified by the important
differential of revenue between these two groups. We could also think that
employed people tend to take more care of themselves as they have social
interactions whereas unemployed people might be sloppy.
Medical visits. The corresponding odds ratio indicates that the person who goes more than 3 times for medical visit is 2.334 times more likely
to be in poor health relatively to the person who goes for less than 3 times.
This result seems to be logical because the person who goes more than
three times for medical visits means that the person is sick as healthy people just take a check-up once or twice a year, so in poor health. However,
not going for medical visits can increase the risk of affecting individual’s
health as individuals can not know if they have an illness if they do not
have at least one annual check up.
Ownership status. According to the results, a tenant is 1.824 times
more likely to be in poor health than an owner. On the one hand, this
can be explained by the wealth of the household, indeed the wealthier are
more likely to be able to pay for health care and this leads to a better
health. Moreover, an owner tends to take better care of his home which
means that their environment is cleaner which enhance the chances of
being in good health. On the other hand, concerning the tenants they
have more housing instability than owners this leads to a higher chance of
being in poor health. Plus, as they have to pay a rent every month they
will have less money to spend on health. All things considered, as health
care is very expensive they will have to arbitrate on the way they spend
Weight. The value of the ratio for obesity shows that obese people are
1.603 times more likely to be in poor health than non-obese people. The
ratio seems to fit the general results of obesity over health. Indeed, they
are more subject to employment discriminations and the risk of cardiovascular diseases is higher in this population. Thus, obesity rise the price
of health insurance : as the price of an insurance is correlated to the personal risk of damage, a person more likely to have cardiovascular diseases
should pay a more expensive health insurance. Moreover, the employment
discriminations force obese people to lower their demanded wage to compete with non-discriminated people. Combining those two effects, obese
people generally earn less and have to pay a higher health insurance. Last
but not least, the psychological effect is not to be underestimated as obese
people undergo pervasive stigma during their life (Carr, Friedman, 2005)
Degree. The corresponding ratio shows that an individual who has less
than bachelor’s degree is 2.348 times more likely to be in poor health than
an individual who has more than a bachelor’s degree. Indeed, the more
educated they are the better job they will have. Moreover, once they get
a better degree they have more chances of having good knowledges concerning health. They will be able to understand which insurance to take,
and how much they will have to spend on it.
Thus, a higher degree leads to a better job, this means that the company they are working for will propose them a good private insurance.
Indeed, the salary and the level of education play an important role in
the state of health, a higher salary will let them benefit from a better
insurance and this will have a direct impact on their health.
Plus, the more educated they are, the more they will take care of their
health as they are aware of the consequences. Also, theoretical approaches
as the Wagstaff’s developpement on the Grossman model can be linked
to these results. A high level of education can increase the productivity
of medical services that can create a disparity on health-status between
Moreover, Subramanian, Kim and Kawachi found similar results concerning the impact of education. However, there is a difference between our
approaches. They consider several level of education : middle educated
individuals; low educated individuals. They found that people with a low
level of education are 3.57 time more likely to be in poor health (and 1.90
for middle educated). The composant of education has a huge impact on
the state of health on individuals in the USA.
Ethnicity. We could think that ethnicity would have a major impact
over health through discrimination and difference of income. However,
the odd-ratio for ethnicity is one of the lower with a value of 1.280. According to Williams D.R. (1996), the ethnic effect over health comes from
the lack of education about the good behaviour, the psychological impact
of discrimination and the exposure to environmental risk factors. Indeed,
giving that physicians have the duty to heal without social nor racial
considerations, it seems correct to think that discriminations have little
impact over health like it was found in Subramanian, Kim and Kawachi’s
article. However, population in the United States of America are generally clustered by their ethnicity and not white people live in the poor
neighbourhoods, so that they suffer from a worse environment which is
Gender. The parameter estimated for the gender is not significant. It
tends to prove that health does not discriminate according to the gender
which is a good thing giving that a health system should be egalitarian.
The results globally correspond to the ones expected in their effect and
they are close to the ones in the Subramanian, Kim and Kawachi’s study
for sex, race and education. However, we can discuss the robustness of
the results because of the data set construction and the selected variables.
We will eventually suggest ways to improve the study.
The theoretical cross-section constraint
The data from the CHIS 2016 survey are collected in cross-section. That
is to say that there is an impossibility to capture the individual heterogeneity like the strength of the immune system.
Moreover, we miss control variables. Indeed, this represents an issue,
particularly for the case of the USA with the introduction of the public
insurance Obama Care 10 , willing to help poor population. For instance,
we we miss a dummy variable to know if the person benefit from the
Obama Care. That causes biases during the estimation as the effect of the
Obama Care might be captured by the variable unemployment, as they are
likely to be correlated. This is to say that we may have found estimates
that are lower than the real one. Another fact to point out is that the
survey takes place in California. Because of this, the results found can’t
be applied to other countries because Americans have their own lifestyle,
environnement and reglementations that will impact health but won’t be
seen in the data selected.
The problem of the outcome measure
The chosen health index is subjective so that it doesn’t measure the physical health state of an individual but a multidimensional representation
which include psychological health and the perception of one physical
ability and disability. Thus, a one-arm person who has accept its handicap could consider himself in better health than a person who recently
starts feeling a chronic pain in the knee. And people with close health
state could represent a very disparate population. Moreover, we created
a dichotomic variable to measure health whereas it was a 5-class variable
in the survey. All those transformations could lead to a mismeasure of
health and, doing so, create a bias in the estimates.
The endogeneity is a major problem in our model. For instance, variables like the weight are clearly correlated to the residuals. Indeed, the
residuals contain information on the metabolism which influences one’s
weight. The Degree variable is endogenous because the capability of the
individual can explain one’s ability to get a degree, plus parents play an
in 2010 and implemented in 2014.
important role as they transmit knowledge to their children and they encourage them to get a higher degree. In general, the more parents are
educated the more children will be educated too. We can say that there is
a kind of transmission of genetic and cultural heritage. Furthermore, the
Medical Visits variable is also endogenous, indeed the endogeneity can be
explained by the area. The location one’s lives in can explain the number
of medical visits. In fact, people living in the city have more chances to
find a doctor as doctors are more established in the cities. However, the
psychology of the individuals can affect their number of medical visits,
indeed, an hypochondriac person will tend to go more often to the doctor.
The Ownership status is also endogenous. The revenue of the individual
and probably its inheritance influence if one is a tenant or an owner.
In order to find better results, we thought about various econometrics
possibilities. At first, we could have made a Panel Model which would
have corrected the individual heterogeneity. Likewise, a Probit Model,
which is a binomial regression model, would have allowed us to instrument
our variables whereas the Logit Model does not allow us. Also, our model
is based on a binary dependant variable, thus there is only two state of
health. We could have use a multinomial logistic model in order to have
a polythomic dependant variable that could explain several possibilities
of state of health.
This study of the social determinant of health seems to confirm the results
of Subramanian, Kim and Kawachi. Thus, we can think that the estimates
found are realistic and close to the real one. Another argument is the correlation with the pyramid health impact Thomas R Frieden (2010) . As
in his framework, we found that actions requiring an important personal
investment increase the chances of being in good health more than large
actions requiring little personal investment. Indeed, the odds-ratio for
employment, education and medical visits which are the variables that
require the most personal investment are higher than the others which
seem to represent large scale actions that require less individual effort.
Freudenberg and Ruglis (2007) insists more on the education component.
For them, it is clear that the education level can predict the state of health
of individuals. The Wagstaff’s extension (1986) of the Grossman model
goes in the same way. In the case of the californian State, these intuitions
are confirmed by our results.
Individuals are unequal concerning health state. Especially in a country
like the USA where socioeconomic inequalities are predominants. Despite
the will of the US government to organize some assistance for poor individuals as medicaid and medicare programs. They fail to reduce these
health inequalities, and the high cost of healthcare services does not help
as it forces a part of the population to depend entirely on their environment and habits to be in good health.
In order to improve the overall state of health, it would be wise to act
upstream with emphasis on education. The government must encourage
young people to be more educated. Indeed, individuals with a better degree are generally better informed about the good habits to stay in good
health, and the consequences of being in poor health. Thus, rising the
minimum age before dropout or fixing a minimum degree could be a way
to increase the overall state of health, especially in area where the graduation rate for high school is lower than 60 percent (CB Swanson, 2004) .
In addition, downstream campaigns need to be put in place to educate
individuals about the importance of health care. As part of the diseases
spread through large population because people don’t know the rudimentary gestures that reduce propagation, it makes sense to think that better
education about hygiene rules would cut out disease propagation rate and
increase the general health state.
 Thomas R. Frieden. A framework for public health action: The
health impact pyramid. Journal of Health and Social Behavior, April
 Deborah Carr ; Michael A. Friedman. Is obesity stigmatizing? body
weight, perceived discrimination, and psychological well-being in the
united states. Journal of Health and Social Behavior, September
 Michael Grossman. The human capital model of the demand for
 Caroline Despres ; Paul Dourgnon ; Romain Fantin ; Florence Jusot.
Le renoncement aux soins pour raisons financi`eres : une approche
´econom´etrique. Questions d’´economie de la sant´e, Novembre 2011.
 S V Subramanian ; Daniel Kim ; Ichiro Kawachi. Covariation in
the socioeconomic determinants of self rated health and hapiness : a
multivariate multilevel analysis of individuals and comunities in the
usa. J Epidemiol Comunity Health, 2005.
 Peter Zweiffel ; Friedrich Breyer ; Mathias Kifmann. Health Economics. Second Edition. 2009.
 F. Galvis-Narinos ; A. Mont´elimard. Le syst`eme de sant´e des etatsunis. Pratiques et Organisation des Soins, pages 309–315, Novembre
 Nicholas Freudenberg ; Jessica Ruglis. Reframing school dropout
as a public health issue. Preventing chronic disease, Public health
research, pratice and policy, October 2007.
 R. Auster ; I Leveson ; D. Saracheck. The production of health, an
explanatory study. Journal of Human Ressources, 1969.
 CB Swanson. Who graduates ? who doesn’t ? a statistical portrait
of public high school graduation, class of 2001. The Urban Insitute,
 A. Wagstaff. The demand for health: an empirical reformulation of
the grossman model. 1993.